The_Decryptor

If one day clang of gcc ships with a local LLM to lint code for security problems I would probably run it occasionally even on my hobby code despite being an AI hater.

Compilers already have deterministic linters built in, devs often turn them off anyway.

Quite a lot of the bugs being found in existing codebases show up as soon as you re-enable them.

And it’s also not like the hardware is improving much either, we’ve effectively hit the wall performance wise with processors and RAM.

That’s literally what all these datacenters are for, we can’t scale the individual computer performance up, so instead we build more and more of them.

The LLMs are wicked fast at reviewing code

That’s because it’s not actually reviewing the code, rather just producing text that statistically looks like review comments.

So it does look vaguely useful, e.g. if code has null checks, review comments usually won’t ask for them to be added, but it’s never actually reviewing, or understanding, or reasoning, or anything else people claim.

It’s like an actor playing a doctor in a TV show, they’re not actually diagnosing patients, they’re just following a script that looks like it.

“Prompting” isn’t a skill, that’s just asking for what you want. Same way as asking a barista for a drink doesn’t require the same level of skill as making it.

But in general yes that’s how arguments and decision-making works, for any law. These things aren’t mathematical proofs interpreted by computers, they’re people you convince of your point.

Ehh, it seems pretty targeted to me, it’s defined to only apply to automatic apps or services, e.g. it specifically exempts manually created images.

Bans X from allowing grok to make naked pictures of celebrities but doesn’t apply to people using photoshop to do the same.